Guide · Updated September 2026
Most guides stop at "count the positive mentions." That gives you a number that looks precise and tells you almost nothing about what to fix. This guide covers the formula, the four ways a single sentiment score misleads a store, what to split it by, how to check whether your tool is getting it right, and a 20-minute weekly routine.
We make Mentionfront, a social listening tool for ecommerce brands. The method here works with any tool, or a spreadsheet.
Brand sentiment is the overall attitude, positive, negative or neutral, that people express toward your brand and products in reviews, social posts and comments. You measure it by collecting mentions, labelling each one, and then comparing the share of positive against negative over time, against competitors, and per product.
The measuring is the easy half. What makes sentiment useful to a store is what you split it by: a healthy overall score can sit on top of one failing product, and a "neutral" post can be a shopper about to buy. Split by product, platform and topic, and read the posts behind any number that moves.
Curious what your own numbers look like? The free social listening scan labels your top 30 mentions from the last 30 days.
Two numbers cover almost every use. Pull the mentions for a fixed period, label each as positive, negative or neutral, then work out:
Positive share = positive mentions ÷ total mentions × 100
Net sentiment = (positive − negative) ÷ total mentions × 100
Positive share is the figure most guides use. Net sentiment is more honest, because it counts the negatives too: a brand with 56% positive and 9% negative reads very differently from one with 56% positive and 40% negative, though both have the same positive share.
Neither has a universal "good" score, because it depends on what you sell and where people talk. Use yours as a baseline, then compare it to itself month to month and to competitors you track the same way.
| Example | Posts | Pos. / neg. | Net |
|---|---|---|---|
| Whole brand | 1,000 | 560 / 90 | +47 |
| Serum | 500 | 350 / 20 | +66 |
| Cleanser | 380 | 190 / 20 | +45 |
| Night balm | 120 | 20 / 50 | −25 |
Made-up 30-day numbers (posts, then positive and negative) to show the shape of the problem: +47 overall looks healthy, while one product is net negative. The three products sum to the brand row.
Questions carry no emotion, so they're labelled neutral. "Does the linen set come in king?" is a shopper with a card in hand. If neutral is 60% of your mentions, most of what you'd want to act on is inside it. Sentiment measures tone; it doesn't say what someone wants.
A brand-level score averages your bestseller with your problem child. The table above shows it: the balm is losing while the brand looks fine. Sentiment is only actionable at the level of something you can change, which is usually a product.
"Love the fit, but it took three weeks to arrive" is praise for the product and a complaint about shipping. One label loses half of it. A theme tag (fit, shipping, packaging) keeps both, and tells you who owns the fix.
Ten mentions a week means a single angry post moves your score by ten points. Look at a rolling four-week window, and don't react to a movement until you've read the posts behind it.
Each split answers a different question and points to a different owner.
| Split by | Question it answers | Who acts |
|---|---|---|
| Product | Which item is losing or gaining people? | Product, merchandising |
| Topic (shipping, sizing, packaging, price…) | What are people actually complaining or raving about? | Ops, CX, product |
| Platform | Is one audience harsher than another? TikTok comments read differently from Reddit threads. | Social, community |
| Intent (buying, switching, complaining) | Which mentions need a reply, and which are just mood? | Sales, social, CX |
| Competitor | Is your gap to them widening? Track them the same way. | Strategy, brand |
| Time | Did it move after a launch, a price change or a bad batch? | Whoever owns the change |
Intent and topic are what turn sentiment from a mood into a to-do list. Mentionfront labels both on every mention; the sentiment and intent analysis page shows how, and product-level tracking covers the per-product split.
Verbatim Hacker News comments from practitioners, each dated and linked to the original.
… I've come to the conclusion that sentiment as a product is useless. There's no actionable conclusion from sentiment alone. You need to dig deeper, apply part of speech tagging, and determine the underlying intent.
Labelling individual sentences without context is hard enough, but what makes it worse is that it then spreads to the mistaken assumption that sentences can be analysed without context based on the initial training.
Sentiment analysis has always been a bit dodgy for individuals, but it mostly works due to sheer numbers. Sarcasm, for example, works very poorly with SA…
… I found that it was impossible to accurately infer 'sentiment' of a single tweet or person (not just because of sarcasm and other nuances). However, when you take the average of a group (wisdom of the crowd) then the results are much more promising.
… we were performing a rudimentary analysis on all non-rated mentions (Twitter, Facebook, etc.), however the algorithm we were using was providing too many wrong sentiments, so we removed it for the time being. …
… we've found that sampling and human analysis/tagging is far more accurate at judging the sentiment around a brand, company or topic.
Most of these are years old, from before language-model classifiers, and accuracy figures in comments like these have moved on, so we left the old percentages out. The advice hasn't aged: judge sentiment in context, expect it to be shaky on single posts and steadier in aggregate, and pair it with something you can act on.
Every tool that scores sentiment will get some posts wrong, especially jokes, sarcasm and mixed opinions. The fix isn't to distrust the number; it's to know your own error rate. This check works on any tool and on any export.
Not the top 50 or the newest. Random, across products and platforms, so easy cases don't dominate.
Positive, negative or neutral, judged toward your brand, before looking at the tool's label.
Agreement over 50 gives a rough accuracy. More useful is the pattern in the misses: sarcasm, product names that double as words, or complaints labelled neutral.
Read ten "neutral" mentions. If several are buying questions or quiet complaints, your neutral bucket is hiding work, and you need an intent label, not a better sentiment score.
Why sample by hand
A sample of 50 is small, so treat the result as a smoke test, not a benchmark. If your tool agrees with you on most of them and misses in a predictable way, you can trust its trend lines. If it misses at random, don't build reporting on it.
A short list of what each labels and what it costs, from vendor pages read on September 29, 2026. For wider tool lists, see the brand monitoring tools comparison.
| Tool | What it does with sentiment | Price |
|---|---|---|
| Hootsuite's free sentiment tool | Enter a brand, topic or creator and it sorts mentions into positive, negative or neutral. It carries a notice that it may show inaccurate material. Useful as a quick check, not a tracker. | Free |
| Determ | AI sentiment and summaries on every plan, across news and social sources. | From €99/mo |
| Brand24 | Sentiment, emotions and topic analysis on general web and social mentions. | From $199/mo billed annually |
| Sprout Social | Sentiment features sit on the Advanced plan, alongside its spike alerts. | $399/seat/mo billed annually for Advanced |
| Mentionfront | Three sentiment levels read by a language model per mention and judged toward your brand, plus intent, an ecommerce topic and a priority score. Covers six social platforms, not news. | From $29/mo billed annually |
A spreadsheet works too, at low volume: export mentions, label a sample by hand as in the audit above, and calculate the two formulas. The point of a tool is doing it continuously and per product, not doing something a spreadsheet can't.
The point is to catch a slide early and connect it to a cause. Same time each week, same order.
Compare each product's net sentiment with its own last four weeks, not with the brand average. Flag any that moved more than you'd expect from normal swings.
Open the ten most negative mentions for each flagged product, ordered by reach. Numbers tell you where to look; the posts tell you what happened.
Tag the reasons: shipping, sizing, packaging, quality, price. Two or three complaints on the same topic in a week is a pattern worth passing on.
Send the specific posts to whoever owns the topic, and write one line: what moved, why, who has it. Next week you'll see whether it worked.